Finite Element Implementation of ANN-Based Constitutive Models for DSA in SS304: Fid-Bau Portal

Finite Element Implementation of ANN-Based Constitutive Models for DSA in SS304

Patra, Siddhartha / Dhar, Sankar / Acharyya, Sanjib Kumar

Stainless steels show a complicated material behavior involving viscoplasticity and Dynamic Strain Aging or DSA, wherein in certain loading conditions, the material exhibits negative strain rate sensitivity (SRS) while exhibiting positive SRS in others. Most conventional models used for modeling DSA lack either the accuracy or the ability to efficiently incorporate the temperature dependence or both. There are ANN and other machine learning-based model which are reported to predict the DSA stresses with reasonable accuracies for a wide range of loading conditions. However, they are not often implemented in an FE formulation. In fact, some authors concluded that at their present stages of ANN modelling, FE implementation remained a challenge. This creates a gap between models which can make reasonably accurate predictions and the models which are available for Finite Element (FE) implementation. The objective of the study was to focus on the application of ANN in modeling of DSA and implementation of the same in FE formulation. Here an ANN-based model was developed for 304 stainless steel. For this purpose, an ANN with two hidden layers was used. The ANN was trained by dropout method and produced better results than a single hidden layer model. The architecture of the ANN most suitable for the purpose was 3-10-10-1. The ANN-based model was implemented through a FE software. Results of the FE simulations show agreement with the experimental observations. Unlike most DSA models both conventional and ANN based, which are often limited by their application either at a specific temperature or in a small range of temperatures, the ANN model presented here, could predict the behavior of the material within and outside of the DSA regime.